Traditionally, mine detection using sonar imagery has used image processing and detection methods that often rely on preset segmentation and predetermined feature extraction methods followed by classification methods (e.g., neural networks, Bayesian networks and so forth). These classification methods have the disadvantage of not providing good detection or classification accuracy.